The Man Who Lived In The Gym

# The Quota Bet Sheet

The Quota Bet Sheet is a free pack of five short markdown files (a decision tree, a two-week tally sheet, a limits grid, a habits list and a same-price comparison) for working out which AI subscription tier is actually yours, using your own usage instead of a pricing page. Comment QUOTA on the reel, or download it below. No email, no signup.

## The story that explains the pricing

In July 2025, Anthropic announced weekly rate limits for Claude Pro and Max. The announcement is still up, and the thread under it is more interesting than the announcement:

> Some of the biggest Claude Code fans are running it continuously in the background, 24/7.
>
> These uses are remarkable and we want to enable them. But a few outlying cases are very costly to support. For example, one user consumed tens of thousands in model usage on a $200 plan.

One person, running a supercomputer around the clock, for the price of a phone bill. The company estimated the new limits would affect "less than 5% of subscribers."

That is not a scandal. It is the pricing model, stated out loud by the company that runs it. And it is the same model your gym uses.

A gym sells everyone unlimited access because most members hardly ever show up. The maths only works on no-shows. The people who go twice a year are quietly paying for the people who go five times a week, and everyone is happy with the arrangement right up until somebody moves in.

Your $20 AI plan exists on the same assumption. You ask a few questions and leave. The light users cover the heavy ones.

So when a handful of people started living in the gym, the walls went up. They are walls no gym ever built, because a gym cannot meter you. Your usage now resets every five hours. And on top of that sits a second, separate weekly cap. Not a replacement for the five-hour window: an additional one, stacked.

Here is the gap the pricing page will never close: it tells you the price, and the price is not the thing you need to know.

Usage is measured in tokens, not messages. Anthropic's own documentation says what you spend depends on "the length of your current conversation, the length of files you attach, and the model or feature you use." One long document can cost more than fifty short questions. Two people on the same $20 plan can have completely different experiences of it, and neither of them could have predicted which they'd get.

Which is why Anthropic declines to publish a per-window message count at all. You will find confident numbers elsewhere: roughly 10 prompts on free, 45 on Pro, 225 on Max 5x. None of those are published by anyone. They are individual creators' personal testing, true for one person on one day doing one kind of work. Useful for a rough sense of the ratios; useless for deciding whether *you* will get stopped.

The only number that decides which plan is yours is how often you personally get stopped mid-task. Nobody publishes that, because nobody can. You measure it, and it takes fourteen days and no effort.

Each day, write down how many times you were stopped mid-task. Not times you were aware a limit exists, and not times you finished and moved on. Stopped, with the thing half-done. Next to it, mark which wall it was: the five-hour session window, or the weekly cap.

Then read the fortnight:

• Zero to two stops: you are nowhere near the edge. Don't upgrade. If you're already paying, run the downgrade check.
• A few, all session-window: that is the most fixable case. Start a new chat when the subject changes, paste the section rather than the whole document, and put heavy work at the *start* of a window. Re-count in two weeks before spending anything.
• Any weekly-cap hits, two weeks running: habits will not move that one. Go up a tier.

And then the rule that matters more than the tree: buy one size above where you actually get stopped, not two. Every tier keeps both walls. Paying more raises the ceiling; it does not remove it.

Set one reminder for three months after any upgrade, and answer one question: in the last three consecutive weeks, did you hit the cap of the tier *below* the one you're on?

If no, downgrade. You're paying for headroom you aren't using, which is exactly the arrangement that makes an unlimited tier profitable.

Upgrading is a decision people make deliberately. Downgrading is one nobody ever revisits, so the subscription quietly outlives the reason it was bought.

Five files, all short, each stamped with its own date because prices drift and these pages get reshared separately:

• `quota-decision-tree.md`: four yes/no questions that route you to Free, Pro, Max 5x or Max 20x, plus the one rule that beats the tree.
• `the-two-week-count.md`: the tally sheet, how to read your fourteen numbers, and the downgrade check.
• `limits-cheat-sheet.md`: every tier's price, session reset, weekly cap and context window in one grid, with VERIFIED or ESTIMATE marked on every single row.
• `free-tier-maximizer.md`: six habits that stretch any tier, and the test that tells a session wall from a weekly wall.
• `chatgpt-vs-claude-at-20.md`: the same-price comparison, with its weaker sourcing stated at the top rather than buried.

Prices and limits are as of 2026-08-15, taken from claude.com/pricing and Anthropic's support pages. Consumer AI pricing moves fast enough that you should check it yourself before acting on anything above.

One thing is marked differently on purpose. The ChatGPT comparison figures are aggregator consensus, not vendor-quoted. OpenAI's pricing pages returned HTTP 403 to automated fetching when this was built, so several independent pricing trackers were used instead. That is reasonable corroboration, not the same bar as the Anthropic numbers, and the file says so before it shows you the table.

The method outlasts all of it. Count what you actually hit, buy one size above it, check again in three months. That will still be right when every number here is stale.

Comment QUOTA on the reel and I'll send it over, or just download it above.